When we validate an analytical determination method with instruments like HPLC, LC-MS etc we conduct the linearity tests to create calibration curves, then we will get the linear equation Y=mx±c. for the intercept here some times we get very low or very high values, what dose the observation result from? Does it relate to the concentration of the analyte? if so how?
Many times we had the same problem. There is even the possibility to get negative concentrations. Most times we do know that for analite concentration of zero the system response should be zero. Our solution is to oblige the calibration straight line to pass by zero; the common spreadsheet programs fortunately have that possibility. In our case it was the best solution we could find, Regards,
Dear yibadatiha
Observation of intercept (high value) was showed that blank measurements is required specially if detection system of chromatography instrument was spectroscopy instrument.
Many times we had the same problem. There is even the possibility to get negative concentrations. Most times we do know that for analite concentration of zero the system response should be zero. Our solution is to oblige the calibration straight line to pass by zero; the common spreadsheet programs fortunately have that possibility. In our case it was the best solution we could find, Regards,
Intercept point at any calibration curve is not expected to be zero or near if it (this is the exception rather than the rule). The fact that this point is not zero is strongly related with the equipment, devices, mainteinance, experiment conditions, baseline/background and maybe other specific details.
One of the main problems might be differences in matrix.
I don't know what matrix you use, but make sure the calibration lines are in a similar matrix as the samples to measure.
Also, if you are measuring a endogenous compound, the intercept will be higher or lower depending on the concentration of the endogenous compound present in the calibration samples.
Speaking in general, I'd agree, as Jorge said, that you use regression through the origin if one would expect y to approach zero as x approaches zero. I suppose, as I think Victor indicated, a non-zero intercept may indicate other problems. But perhaps the real problem, or part of it, may be heteroscedasticity, as noted in the paper by Chris Lee, which I have attached below.
Anothet thing, aren't these calibrations done with very few observations? One 'bad' one could really throw this off.
Article Calibration uncertainty in pharmaceutical analysis: replicat...
Yibadatihan - Hi -
There are some answers I put on the question link below that may be of interest.
Cheers - Jim
https://www.researchgate.net/post/Does_anyone_know_how_to_calculate_the_standard_deviation_of_y_intercept_of_the_calibration_curve
If you do want to see how your points fit into a classical ratio estimator model, you could plot confidence bounds, assuming normality of residuals (which could be problematic near the origin), by using the attached spreadsheet.
Data CRE Prediction 'Bounds' and Graphs Example for Section 4 of ...
- Badges
- Science method
- Similar topics
- Methodology